Human Activity Sensing

Abstract

This paper seeks to capture user data by means of readily available mobile devices such as smartphones and smartwatches to provide inferences with regard to their current activities and trends regarding the same. Because of the unobtrusive nature of the devices being used, it allows for possible applications in the healthcare field as a passive process to detect the onset of various diseases such as dementia. There has been some research in the field with regard to the same which discusses how the changes in a persons physical movements like stride length point towards the onset of certain diseases, and we will exploring the accuracy of the same or nonetheless establishing the groundwork for the same. We will also be constructing a real-time activity sensing platform with a modular design approach. We will use WebSockets to visualise the data in real-time, so the user can also physically see the different patterns which are being classified by the algorithm. The paper also discusses existing methodologies with regard to Human Activity Recognition, mainly using prevalent classification algorithms such as SVMs and Decision Trees, however as we are trying to pattern user activity along with the attempt to actually detect the onset of certain diseases, Deep Learning models seem more adept at the task. Thus, we will also be comparing the results we get with that of pre-existing classifiers. Finally, we will be listing out the different activities being detected by the classifier, and compare their results with state of the art systems. We will also be looking at a few special activities and movement patterns, such as stride lengths, in order to observe whether they can be used to accurately predict the onset of certain diseases.